Research map / Architectures and efficiency / RNNs and normalization
Deep Multi-task Representation Learning: A Tensor Factorisation Approach
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use DMTRL-TT instead of STL | MNIST 1-vs-rest · MNIST 1-vs-rest, 1% training data | binary classification error rate (%): 11.52 -> 7.31 | Helped |
| Use DMTRL-TT instead of the best classic MTL with CNN features | MNIST 1-vs-rest · MNIST 1-vs-rest, 1% training data | binary classification error rate (%): 9.41 -> 7.31 | Helped |
| Use CNN features instead of PCA features for classic MTL | Classic MTL · MNIST 1-vs-rest, 1% training data | binary classification error rate (%): 14.08 -> 9.41 | Helped |
| Use soft, end-to-end sharing in place of manually selected hard sharing | Deep multi-task learning · MNIST 1-vs-rest, small training set | binary classification error rate | Helped |
| Use DMTRL-Tucker for gender classification rather than STL or UD-MTL | AdienceFaces · Gender classification | error rate | Helped |
| Use DMTRL for age-group classification rather than STL | AdienceFaces · Age-group classification | error rate | Helped |
| Use hard-sharing UD-MTL rather than STL | AdienceFaces · Age-group classification with larger training sets | error rate | Hurt |
| Use DMTRL rather than STL | Omniglot multi-alphabet recognition · Omniglot, all reported training fractions | average error rate across tasks | Helped |
| Use hard-sharing UD-MTL rather than STL | Omniglot multi-alphabet recognition · Omniglot, very small training fractions | average error rate across tasks | Hurt |
| Use hard-sharing UD-MTL rather than STL | Omniglot multi-alphabet recognition · Omniglot, training fraction larger than 50% | average error rate across tasks | Helped |
| Increase UD-MTL hidden neurons to bring its parameter count closer to DMTRL | UD-MTL · MNIST 1-vs-rest, 1% training data | error rate (%): 9.34 -> 9.39 | No clear effect |
Closest papers by meaning
- Learning Sparse Sharing Architectures for Multiple Tasks (AAAI 2020 · similarity 0.88)
- Latent Multi-Task Architecture Learning (AAAI 2019 · similarity 0.87)
- Near-Optimal and Efficient First-Order Algorithm for Multi-Task Learning with Shared Linear Representation (ICML 2026 · similarity 0.87)
- Sharing Knowledge in Multi-Task Deep Reinforcement Learning (ICLR 2020 · similarity 0.87)
- Provable Multi-Task Representation Learning by Two-Layer ReLU Neural Networks (ICML 2024 · similarity 0.84)
- Learning Multi-Task Sparse Representation Based on Fisher Information (AAAI 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.